AI for Comprehension: Sentry Engineer's Workflow

Priscila Andre de Oliveira, a Senior Software Engineer at Sentry, shared her insights on how she utilizes AI to enhance her daily workflow, emphasizing the crucial role of comprehension in software development. In a presentation titled "Comprehend First, Code Later: The AI Skill I Rely On Daily," Oliveira highlighted that the most significant impact of AI in her work is not in generating new code, but in understanding existing codebases.

From Senior Engineer to Agent Manager

Oliveira, who has been with Sentry for six years, has transitioned from a traditional Senior Software Engineer role to an "Agent Manager." She humorously noted that while this promotion didn't come with a salary raise, her "reports" (AI agents) don't complain. Her personal experience with AI, particularly with the Claude model, has led her to rely heavily on it for understanding complex code. She revealed that a significant portion of her prompts, around 67%, are focused on comprehension, with only a small fraction (2%) dedicated to actual code generation.

The "Catch Me Up" Skill for AI-Assisted Comprehension

Due to the repetitive nature of her questions to AI regarding code comprehension, Oliveira developed a custom skill named "Catch Me Up." This skill, stored locally on her machine, structures prompts into six distinct exploration modes: Architecture, Convention, Feature Trace, Syntax/API, Testing, and History. This structured approach allows her to efficiently query the AI for context, such as understanding code structure, patterns, or the reasoning behind specific changes. She demonstrated how this skill helps her gain a deeper understanding of unfamiliar code, enabling her to contribute more effectively.

The Bottleneck of Understanding, Not Typing

Referencing a blog post by Jacky Natson, "Vibe Coding Our Way to Disaster," Oliveira echoed the sentiment that the true bottleneck in software development is not typing code, but understanding it. Natson's three-phase methodology of Research, Planning, and Implementation is complemented by Oliveira's emphasis on the AI agent's role in research, followed by human comprehension. She believes that the ability to understand what the AI has found is paramount. This aligns with her personal workflow, where she prioritizes comprehension before implementation, allowing the code to follow naturally.

The data from her own usage analysis, showing a high percentage of comprehension-focused prompts, reinforces her belief that AI's value lies in its ability to help developers navigate and understand vast and complex codebases more efficiently. She highlighted that AI serves as a tireless teammate, capable of answering questions without getting tired, making it an invaluable tool for developers.

Embracing AI for Faster, Deeper Understanding

Oliveira concluded by encouraging developers to leverage AI for comprehension, emphasizing that the AI is a tool to augment, not replace, human understanding. By using AI to explore and understand code, developers can significantly speed up their workflow and ensure the quality of their contributions. She also shared that Sentry itself is embracing AI across its platform, with tools like "Abacus" for AI coding tool usage, "Warden" for AI-powered code review, "Junior" for bug detection, and "AI SDK Testing" for integration checks, all contributing to a more efficient and intelligent development process.